I disagree, you could do a graduated wallet, or you could have an LSP that locks up less liquidity and charges comparatively higher fees. Of course in that case the LSP would never need to charge more than on-chain - you still get *some* “compression”.

Then you're not really criticizing any particular constraint, are you? You're criticizing lightning itself, but the year is 2026 and I don't see too many other scaling solutions around the corner